Loading summary
Transcript105 lines
- [00:00]
Cigar Bid Representative
Attention cigar enthusiasts. Today we're thrilled to introduce to you Cigar Bid, America's number one online cigar auction site. If you're looking for unbeatable deals on every type of cigar, listen up. At Cigar Bid you have the power to set your own price, often scoring top quality cigars and accessories at prices much lower than retail. Others want you to believe they are always giving you their best deal, but at Cigar Bid you get to decide the deal. It's the cigar site the rest of the industry does not want you to know about. Score deals at the best prices every single day. The live auctions are fast paced and exciting, making every bit a thrilling experience. Check out Freefall to watch the prices drop in real time or the power hour for our fastest pace auctions. Cigar Bid has a text update so you get notified when you're outbid and get exclusive deals right to you. And if bidding is not your thing, then no problem at all because you can buy the cigars straight up at awesome low prices as well.
- [00:54]
Greg Osuri
So.
- [00:54]
Cigar Bid Representative
So no matter what you're looking for, you'll find it at Cigar Bid. So if you're ready to enjoy the fun and the convenient way to purchase fine cigars at great prices online, visit cigar bid.comCrypto101 that's cigar bid.comCrypto101 Happy bidding. Have you ever wanted to trade bitcoin but haven't Dared tried? With Plus500 futures, you can trade crypto without the hassle of opening a wallet. With just a few clicks you can register and start practicing with their free and unlimited demo. See a trading opportunity. You'll be able to trade in just two clicks. Feel ready? You can move to real money with as little as $100 once your account is approved. And the great thing is that in addition to crypto, Plus500 gives you access to a wide range of instruments like the S&P 500, NASDAQ, gas and much more. Explore equity indices like Energy, Metals, Forex and beyond. With a simple and intuitive platform, you can trade anytime, anywhere. Experience the fast accessible futures trading you've been waiting for with +500. With over 20 years of experience, +500 is your gateway to the markets. Visit us. +500.com to learn more. Trading in futures involves risk of loss and is not suitable for everyone. Not all applicants will qualify and this is not personal financial advice. Plus 500 it's trading with a plus.
- [02:15]
Bryce
All right all you good wonderful citizens of Crypt nation. Welcome back to another action packed high caliber episode of the Crypto 101 podcast. I'm your co host, Bryce, as always, joined by my good buddy across the nation, Mr. Brendan Veeman. Brendan, are you excited for this episode today? It's going to be a little different than what we normally do.
- [02:35]
Brendan Veeman
Yeah, it is. You know, we get to talk about all things AI, and that's a giant hot topic right now. In fact, if you're new to the crypto markets, in fact, if you've never heard of the crypto markets, you are still familiar with artificial intelligence. It's the talk of the town. Everyone's interested in it. And what we're interested in is what kind of role does AI play in the crypto markets? So we brought in one of the leaders in the space to discuss it and, man, I'm excited for this episode.
- [03:03]
Bryce
Yeah. Thank you so much, Greg Osuri, who's the CEO and founder of Akash Network, for joining us today. Greg, how are you doing today?
- [03:12]
Greg Osuri
Doing great. Thank you for having me.
- [03:14]
Bryce
Yeah, we're, we're really excited about AI Simply, like, if you're in the audience and you're listening to this and you're not using AI, you're simply behind the curve. It's helped me get such a better understanding of markets and, and all sorts of things. It's helped me be more productive. And so you've seen the whole chat GPT boom, and then, you know, Facebook and Google and everybody comes out with, you know, new AI products. But there's, there's something much deeper going on here. It's not just these chatbots, which a lot of people like, oh, AI, it's just a chatbot. It's been around for years. But there's, there's something really revolutionary here. And we, we want you to join us on this journey here with Greg as we, we learn about the future of AI and really what is it. And so there's a lot of content out there about Greg, you know, that you could learn about, you know, his past and learn about what he's building. But we're really wanting to take a deep dive on, on AI with you today. So if that's cool, Greg, we're going to sk the small talk and we're just going to dive into some of the big questions. Let's do it. What the heck is AI and why is it here? Why is it just on the scene right now? What is it? What is it doing?
- [04:21]
Greg Osuri
It's a loaded question. AI, you know, the way I see it is augmentation of intelligence. For the first time, we are able to extend our ability to think. And if you compare that to the Internet, Internet in my definition is augmentation of reach. Right? So before Internet, your reach was confined to physical boundaries. Post Internet, your reach was basically confined. I mean those boundaries went away and really the network opened up to anyone in the world. With mobile that accelerated. Right. But AI is really augmenting intelligence. And what that means? Well, when intelligence can be commoditized, that means we can improve human productivity at levels that has never been seen before. Now there are two ways to look at it. One way to look at it, is it going to replace human beings or is it going to enhance human beings capability? So I obviously fall into the camp of enhancing our capabilities. Right. And another way to look at AI is it makes smart people smarter, dumb people dumber. Right? Because now in evolutionary standpoint, our brains, human brains are actually getting smaller, you know, because now we don't have to use so much of our brains that we had to use when we were in the savannah. Because savannah, this constant danger, this constant calculations, constant evolution, I mean usage of brains, our brains were much larger. But as humans are getting more specialized and less generalized, our brains are actually getting smaller and smaller. So in a world you can make prediction that hey, we are actually getting, not getting dumber, but we're getting like less sort of like breath when it comes to thinking and more focused thinking.
- [06:34]
Bryce
So people, we're left probably healthy too and less strong. I mean our body, we're much more sedentary and eating processed foods. So that de evolutionary pressure definitely is happening on the human race a lot of different ways.
- [06:46]
Greg Osuri
So one might argue that without AI we may not have a human race any longer. Because now we can offload some of the intelligence that we lost to a machine and how humans have done over the years, all the way from the invention of the wheel. So AI is an evolutionary step for humanity. And I think in 100 years I definitely can see a birth of a new species that's augmenting AI and humans. And that's really what AI in my definition is.
- [07:23]
Shopify Representative
Shopify is a global commerce platform that helps you sell at every stage of your business and sell more with less effort. Thanks to the Shopify Magic, your AI powered all star sign up for a $1 per month trial period at shopify.com redcircle all lowercase go to shopify.com redcircle now to grow your business, no matter what stage you're in. Shopify.com RedCircle you just realized your business needed to hire someone yesterday. How can you find amazing candidates fast? Easy. Just use Indeed. Stop struggling to get your job post seen on other job sites with Indeed sponsored jobs. Your post jumps to the top of the page for your relevant candidates so you can reach the people you want faster. According to Indeed data, sponsored jobs posted directly on indeed have 45% more applications than non sponsored jobs. Don't wait any longer. Speed up your hiring right now with Indeed and listeners of this show will get a $75 sponsored job credit. To get your jobs more visibility@ Indeed.com podkatz12 just go to Indeed.com todkatz12 right now and support our show by saying you heard about Indeed on this podcast. Terms and conditions apply. Hiring Indeed is all you need. At Ritual, we're celebrating all moms this Mother's Day, current and aspiring. So whether you're looking for fertility support, natal Coline, or our best selling clinically backed prenatal, we've got you covered. Find your ritual, go to ritual.com podcast and for a limited time, get 40% off your first month. These statements have not been evaluated by the Food and Drug Administration. This product is not intended to diagnose, treat, cure or prevent any disease. Based on net sales from industry Data, including Nielsen IQ, Stackline Enumerator as of Q4 2020 in top five products.
- [09:18]
Bryce
Wow, that's incredible. And, and I hear so many, like, discussions about AI now in the media. It's, it's everywhere. And it's really all about this infrastructure that it seems like there's a big debate about where is the future of AI really going to be built and hosted and who's going to control it. It almost seems like there's a war unfolding. It's a trade war, right? In a sense. But if you really look into what's going on, it's about, you know, controlling the chips and controlling energy. And so there's a little, and again, a lot of this is, is between China, right? China and America, two of the big superpowers going to head to head on who's going to control the future. So it's a crazy, crazy, Almost Cold War 2.0. I don't know if that's an appropriate way to put it, but I want to kind of put that out there to you, Greg, and just get your whole take on, on what's unfolding right now, right?
- [10:08]
Greg Osuri
So if you break down the different layers of AI at the foundational layer, you have compute. All right. I mean, you can even break down further. I mean, at the Foundational layer. You have the chips, right. And energy. And if you break them down a little further, you have rare earth minerals, essentially to make the chips.
- [10:27]
Bryce
Right, Right.
- [10:28]
Greg Osuri
So you have minerals, you have energy and magic in the middle, then you have these chips, and then you have, you know, the ability to connect these chips and serve AI. Right. So that's the infrastructure level. And beyond that, you have models and data and, you know, coming together along with compute that produces AI. Right. So there are two aspects for AI. There is training aspect that means you're training a machine on how to think, and then there's the inference aspect, which is used in the machine to do what you want it to do. So when we use ChatGPT, we are using the inference of the AI, right. So the training aspect of it requires enormous amounts of chips co located in a single location. So what that means is so today, if you were to see Grok 3, which is the most advanced model, you know, it needs about 35 megawatts of energy, so about 35,000, you know, advanced chips made by Nvidia to run the.
- [11:46]
Bryce
Whole system currently, just to train. Just to train it.
- [11:50]
Greg Osuri
Train.
- [11:51]
Bryce
And is it training constantly?
- [11:54]
Greg Osuri
No, it's periodical. So it takes, you know, six months training, depending on how big the model is. I can't remember how long the GROQ actually took to train, but I believe. But it needed about 35,000 megawatts of capacity. Right. To train. And some models could be trained faster with bigger chips and really depends on the model. So it's not always linear in terms of how it can scale and the amount of energy, which is also translated by the type of chips we need, is doubling every two years. Wow. So the GROQ was 35 megawatt capacity. And I believe the data center that trained GROK is based in Memphis, Tennessee, and that is currently drawing about 150 megawatts installed capacity. That's very, very high. And it's only drawing about 7 megawatts from the grid. So the rest of the energy, they're actually burning fossil to power. LNG to power. Because in America, it's nearly impossible now to get anything over 10 megawatts. So forget 30 megawatts of energy.
- [13:17]
Bryce
For somebody who's listening, like, you know, 150megawatts, you might be like, well, what does that mean to me? Think about 150 megawatts, like powering 150,000 thousand average American homes on a. On a daily basis at the same time. Like, that's, that's a lot of energy.
- [13:33]
Greg Osuri
I mean, maybe more. Yeah, I mean maybe more. I don't, I don't assume a home on average takes a kilowatt per day. Good, good. Yeah. So that's, that's quite a lot. Right? So a lot of energy. The problem with America is our grid hasn't grown much in the last 20 years. I think we grew about 10% because we never needed to grow that much. I mean the population growth hasn't been that great. Even with population growth, think like we've gotten efficient over time with how much energy we use and whatnot. But now there's a new challenge in the grid with AI infrastructure and EV infrastructure coming online and the grid is aging, right? So it's very, very, very hard to build new plants. Now look at all the options that we have and to really power, let's say, you know, I mean 50 megawatts you can get away with solar. I mean solar is also very challenging because it's not amount of, it's not capturing the energy but also storing the energy and transmitting the energy. And these are not very efficient. Storage is extremely expensive. It relies on batteries, right? Lithium ion batteries. So in particular that means they rely on lithium cobalt and for another component, lithium, another rare earth minerals. And these rare earth minerals constitute about 60% of the cost and that is controlled by China. Today when you talk about geopolitical tensions, China is limiting rare earth mineral import. So that's not going to reduce the cost of solar energy. That's going to impact our ability to capture more energy. That's one problem. Second is we don't have. And the next best thing is probably hydro. Hydro is very, very challenging. So hydro can support up to 100 megawatt capacity. So as we keep growing to support 200 megawatt we need coal or a small modular reactor. Small modular reactors. Now SMRs have been in the process. A nuclear reactor. Yeah, SMRs have been in the process for a very long time and doesn't seem like they're going to be anywhere close to being production ready at least by 2030. And we don't know how efficient these are. This has been a pipe dream for a long time. And after that is nuclear, right? So anything about a gigawatt that means thousand megawatts is a nuclear reactor. In America we have about 96 reactors that are utilized, fully utilized 90, 93% of the time. The last remaining reactor was a three mile mile long island in Pennsylvania that was scooped up by Microsoft on a 20 year lease. So we cannot build. And the last reactor we built was in Georgia, which took about 14 years to build. Wow. We cannot build nuclear reactors fast enough to demand. So by 2030 or even sooner, maybe we will run out of energy to train new models. So I will not evolve in terms of how big it can get in terms of models by 2030 or sooner, depending on what. There's a lot of discussion around the energy sector now. There is a $500 million commitment, a billion dollar commitment called the Stargate program by SoftBank and by a bunch of others, I think primarily SoftBank and Microsoft that was promised under the new administration. And most of that capital is diverted towards funding new energy sources and building new data centers. But no one has a clue. I talked to the people on the program. No one has a clue where to get that from. The only options right now is burning coal and lng. And that is challenging because it's dirty. It's literally in Memphis right now. People are complaining about this. Increased cases of respiratory incidence diseases in southern Memphis. And there's a lawsuit that's looming against xai. I don't know what the status of it, but as a few weeks ago they're about to file a lawsuit on AI. Right now is killing people. That's where we are in the stage. And no one's talking about the bitcoin energy anymore.
- [18:09]
Bryce
Yeah, we got a new. A new bad guy in town. Yeah.
- [18:13]
Greg Osuri
So that's one challenge. And now we look at the inference. Right. So this is just training. We're not. We haven't even gotten into inference.
- [18:20]
Bryce
Now inference, like when people are asking like hey chatgpt, how do I do X, Y and Z? Like all of that is the inference correct?
- [18:27]
Greg Osuri
Okay. As we know, I mean the AI is moving so fast and evolving so rapidly. It's having ChatGPT obviously was a big moment for AI when Chat GPT came. People have completely different description of AI. It is the aha moment in a way. All the Transformers has been around for a while. Nobody really cared till ChatGPT came. Ghibli the recent about a few weeks ago. The two, three weeks ago Ghibli moment alone added about a million users to OpenAI. Right. To ChatGPT.
- [18:55]
Bryce
The new image creation stuff.
- [18:57]
Greg Osuri
Correct. I mean it's been around for a long time, but it's just new capabilities that are getting added. And the rate at which AI is improving, self improving in the sense like people are using AI to actually improve AI. Human researchers is rapid to a point. We are predicting at least a good 20% growth year over year. If you look at actually how you look at Internet and cell phones grew, you look at the multiple estimates on how far AI can go. We took a mid estimate. There's very optimistic estimates, There are pessimistic estimates. Right. I mean, this is a whole whole new world, a whole world out there to understand the energy needs. Inference is going extremely fast as inference as AI gets more relevant in everyday world. So today we are looking at what, a billion users spread across Apple intelligence, spread across ChatGPT, Google and different devices using AI and doing about five queries a day, five prompts a day. And that's projected to grow 20% to about 150 prompts a day. And that's very conservative. If you look at people like me that are power users, I write code now, Vibe coding we call this new world where you're literally like using AI to write code. Because as CEO of a company, I don't get too many opportunity to write code because it's time consuming. But now I can literally write code anywhere I want and I don't need the same level of focus and attention that I had before. So people like me would use maybe a thousand queries a day. So we're talking about five queries a day, which is very conservative in terms of estimates. Right. So the AI energy need alone is I believe, growing at around 37% year over year.
- [21:00]
Bryce
Wow.
- [21:01]
Greg Osuri
Right. So now the question is like, how do we solve this problem, how do we solve the energy problem? And if you think AI is going to be center of our existence and so critical to revolution, and especially when you're looking at the rate of progress, Jubilee is just a small aspect, but people are, there's clear product market fit in the developer space, the wipe coding space. Now, I don't even in our own company, if you are someone that's not taking advantage of AI as a programmer, you're going to be left behind. And programmers tend to be the early adopters and usually with lots of tech, be it mobile, be it Internet. Right. So we've seen the same trend as previous tech. And now I can't imagine coding without AI. Like it's like just like you couldn't imagine coding not having the cell phone about 20 years ago. Like 20 years ago you say, well, this is going to be in everybody's pocket, it's going to be a thing. But now you cannot imagine not having the cell phone. Right. Like a smartphone, essentially. It's going to in that trend.
- [22:10]
Brendan Veeman
You know, where do you think that we're like, where will all the Energy come from for us to run society and still progress in AI?
- [22:18]
Bryce
That's the question I've been thinking about this whole time. I'm like, wait, hold on, there's a.
- [22:22]
Brendan Veeman
Where are we going to get all this from?
- [22:23]
Bryce
Where is this coming from?
- [22:25]
Greg Osuri
So lots of people are working on lots of different, you know, lots of different solutions. The most abundant energy on planet Earth is solar sun. My sun hits you six solid hours a day. Solid hours of capture time. Right. I mean, you have other sources, be it wind, bit, you know, hydro and whatnot. But sun is the most abundant because it touches pretty much every place on Earth. Right. And it bugs me that why are we not capturing this energy?
- [23:03]
Bryce
Because it's expensive to store. Right.
- [23:04]
Greg Osuri
Expense, store, expensive, transmit. What if we made it more reliable to store and less, you know, lower the need to transfer the energy? Can we achieve the. Or can we solve the energy problem? Right, we can. So think of a scenario where. Well, today you talk about training being hypercolocated in the sense like you need all these GPUs in a single place. Of course, getting concentrated energy is a challenge. It's not getting energy. We have a lot of energy in the world, we're just not capturing it. Why? Because we actually capture quite a lot. Like in California, for example, there's abundance of energy capture. In fact, during the day we collect a lot more energy than we use. It's just that in the night we need the energy. That's the problem. Right.
- [24:00]
Brendan Veeman
It almost feels like if we can prove that solar is capable of handling something as large and as demanding as AI, that would almost kind of like validify its, its use case for everything else. Right. Because if we can power something that is as intensive as AI is, as, as you've talked about, then that would make me think that the use case for solar empowering so many other parts of society would be there as well, would it not?
- [24:31]
Greg Osuri
Yeah. So solar, for example, is an incredible source. In fact, there's a scale called Kardashau scale. This talks about how advanced a civilization is.
- [24:47]
Bryce
So I've heard Elon talk about that before.
- [24:49]
Greg Osuri
Yeah. So it's a scale goes from 1, 2, 3, and it was by a Soviet physicist astrophysicist in the 60s. And it basically says, you know, type one civilization is a civilization that captured all the energy in its home planet, which is planet Earth. That's naturally occurring. Right. And type two is solar system, type three is galaxy. Right. So if we were to. We are somewhere about 0.7, that means we are not Quite capturing everything we're getting there. If you say we want to advance to one, we have to capture the energy. Solar is the most efficient. But the only challenge is you cannot use that for AI because remember we need co located GPUs, we need GPUs close to each other. That's the nature. But that's changed about six months ago. And that's the key here. There's enormous amount of research, enormous amount of work that went into distributing AI training by labs across the world. Google DeepMind to our labs Pro Alice or News Research or a whole lot of teams working on this problem. It's been around for a long time. Google really released a paper called Dialeco and a bunch of companies that actually implement this paper and actually proved we can train a 32 billion parameter model, fully distributed, not decentralized yet distributed in the sense like you can have a cluster in New York, you can have a cluster in Singapore, you can have a cluster in San Francisco and actually contribute to a training run and actually train a 30 billion parameter model which is close equivalent to GPT 3.5, which is still two years away and two years behind in terms of tech. But it's possible that was not possible a few years ago. And there are companies now like News Research, which I'm very excited about. That means former stability AI. These are extremely respected, well respected teams in the AI field itself. They wrote a paper or released a model called Distro or distributed training on the Internet. Now that made a lot of circles in the space and it definitely seemed to be the more acceptable and more advanced model in terms of can we train them on the Internet? And they're coming up with the testnet I believe in a month or so. And if they can prove we can do 100 billion parameter model, I think that will turn a lot of heads. Right? Yeah, so, so you can see a trend here. If you can actually train using distributed clusters, that means we can have distributed clusters in places that can have about 100 kilowatt capacity. Now how can we get 100 kilowatt capacity? Solar. Right. So a 4,000 square foot roof can get you 100 kilowatt capacity.
- [27:59]
Bryce
So we could just now have, instead of these mega centers of just huge, huge data centers that are burning coal, like whatever, you know, Elon's doing in Memphis, we can now have, you know, millions and millions and millions of these different little nodes essentially collecting solar processing. And hopefully they're going to be doing that because they're earning some kind of income for for doing that. Is that kind of where this whole idea goes?
- [28:25]
Greg Osuri
Yeah, so I mean, so with that's where I think the most exciting thing about distributed network is if you add incentives on top of it. Yeah, some incentive on top of it. And usually incentives form in a share of inference revenues. For example, ChatGPT charges 200amonth now for the most advanced capabilities. Say ChatGPT was trained using this distributed clusters that are across the world and there's some token that you get that guarantees you a share of the income from the revenue that you get from inference. That's a good model. I think that's really where decentralized technologies, be it Deepin or DAI in general is going towards. And a lot of the Jensen is another company that came out recently, I mean with the testnets. Right. So a lot of these decentralized AI companies are now experimenting either through testnets or alphanets with incentivizing compute contributions to train large clusters. And if you combine that with energy markets, there are companies like Daylight Energy for example, right. And say a 16Z Anderson Horowitz funded protocol that announced their round recently and they're doing quite a lot of work in the energy sector. Very interesting model where they use DEFI rails to finance solar clusters. So you're literally connecting DEFI with energy and they can tokenize this debt and service that debt with the earnings. So one of the cool things or interesting things, I wouldn't say cool things, interesting things about solar energy markets is California for example, produces I believe 60 gigawatts of capacity during the day, six hours. And they produce a lot more than they use actually. So there's abundance in California during the day, but no one's charging their EVs and people are in their offices and whatnot. Of course the homes are going to be not utilizing capacity, but in the night when the switch is over, they have to literally fire up all the nuclear plants and the coal plants because energy is real time, there's no storage, it takes in a matter of seconds. They have to fire up all these plants to produce energy or you're going to have blackout. It's kind of crazy if you look at how energy infrastructure, we don't really think about these things, like how mission critical some of these aspects are. So anyway, so if you can store the energy during the day and sell it back to the grid within those seconds you can make a lot of money. And so this company has figured out an algorithm to do so. So they're financing energy grids and if we can Finance small modular data centers. We're talking about 35 kilowatt to 50 kilowatt data centers cost you anywhere from 2 to 3 million dollars. These are not cheap. These are still very highly capable H200 HGX clusters. I mean this is the most advanced GPUs Nvidia has made. If you can power them using the solar panels, you can effectively create an alternative to these hyperscale data centers. And you need about do the math by 2040, you know, with about 2 million of these, we can solve in a global energy problem. Two million is a lot of that. I mean, 2 million times $2 million, right? So it's not cheap. It's like 2% of global GDP or something.
- [32:04]
Brendan Veeman
Yeah, well, it's almost like arbitrage trading in a sense. Like you're finding this disparity when there is the need for power and when it might be transitioning over and you're kind of saying, hey, like I'm going to play the arbitrage here, like, and you find the exact moment in time that this needs to happen, you're going to be like, hey, I have something that you all need. And then they have to essentially pay a premium for it. And what I would compare this to all the crypto traders to is kind of like supply and demand. You know, we see this happen on order books, we see this happen with liquidity and liquidity pools. And when there is demand for something and the supply or the liquidity for it runs out, well then the cost goes up to get your hands on it. Right? Because all these people are saying I want, I want or I need, I need. And there is such a few supply. So that if you can come in, in that moment and say, hey, I can help people capitalize on it, that's where you get kind of the premium in the first place. And you know, it. Yeah, it's unique about this.
- [33:01]
Bryce
And one of the other things that we haven't even talked about is, is bitcoin's demand for energy as well. Like, and I, I've heard people talking about how bitcoin can balance the grid, right? Or it's, it's a, it's a, a load balancer or something like that, which of like we need to bring somebody on to who, who has that sort of thesis who could talk about it, who's like deep in the energy industry and doing, you know, carbon recapture and bitcoin mining and stuff. But Greg, while we had you, I'm curious, like, is there any, like, do you ever think about bitcoin mining and how that might come to bear on. On energy and AI as well.
- [33:36]
Greg Osuri
Yeah, I mean, I. The argument about bitcoin being a load balancer, I would say I would rephrase that as a load consumer, like, because, you know, like I said, like, we have, we produce quite a lot of energy. Right. So one way to look at bitcoin is like, well, look, when you have excess energy during the day, I mean, this is all public information, by the way. There are apps. You can see how much energy Texas produces, California produces. You can literally see during the day, so much excess. Right. In Texas, for example, we have enormous bitcoin mines, Right. That are consuming this excess energy, the solar panels. I mean, Texas, I believe, has got second largest deployment for solar panels in the country after California. So it's great, great way. I mean, you have to burn that energy otherwise. Right. Somehow. So instead of burning it, mine bitcoin and use the bitcoins that you earn to buy back energy when you need it. Perfect. Makes sense from a, from a load balancing standpoint. Right, so, and there's also like thermodynamics, you can say, like at the. The energy that the miners produce can be rotated back into producing more power. So there's quite a lot of interesting things you can do because as long as you see thermodynamics play out, you can always optimize the flow. But bitcoin energy consumption is minuscule because.
- [35:09]
Bryce
Minuscule compared to AI is what you're saying. Yeah, yeah.
- [35:13]
Greg Osuri
Okay. I think right now, I mean, I don't have the numbers on top of my head. I think right now bitcoin consumes 90 gigawatt, 19 gigawatts of capacity. That's about 19 nuclear reactors. You can literally power bitcoin. Right. The network, it's bigger than Sweden, bigger than some countries in terms of energy consumption. But in the grand scheme of things, it's not that much compared to AI consumption. AI consumption will be in, I believe, 8% of global energy by 2040.
- [35:43]
Brendan Veeman
Why is it then that people have come after bitcoin and the bitcoin miners? And I've seen articles about it for years about why people. Yeah. And then, like, I don't see the same kind of, of, I don't know, flack towards the AI sector. If it's so much more like, why doesn't I get the same level of like, flack or hate or whatever we want to call it?
- [36:06]
Greg Osuri
Because where do you think most of the AI is built now? China? No. San Francisco.
- [36:15]
Bryce
San Francisco.
- [36:16]
Greg Osuri
What political spectrum do San Franciscans fall under?
- [36:21]
Bryce
So they're, they're benefiting from all of this AI growth, but they're not benefiting from the, the crypto growth. So they're going to talk their. Talk their book basically, and shit talk the other guys. Yeah, that makes. That's a tale as old as time.
- [36:32]
Greg Osuri
What's funny is no one is talking about the Memphis health crisis, even though Elon is probably one of the most hated people in the country now, especially by the left. But people are very surprised how little news coverage is. I mean, there's some articles, you can Google it and see XAI and asthma and whatnot. You'll get like a few articles here and there.
- [36:56]
Bryce
Yeah, I hadn't heard about it until you, until you mentioned.
- [36:59]
Greg Osuri
Yeah, no one heard about it. It's a big deal. Like, so, you know, the other thing.
- [37:03]
Bryce
Kind of on this whole like, you know, let's call it downsides of AI or whatever, just like the, the dark side or things that people just don't typically talk about. I want to ask you about something that's always confused me is AI safety and alignment. It's like, I don't even know what alignment means in this way, but I know what safety. It's like, hey, making sure that the AIs don't kill people or subvert people. But how do you think about AI safety and alignment? And can you kind of define those things for us here?
- [37:31]
Greg Osuri
Yeah, so alignment argument is so chat GPT, right? It's closed system like most AI today. That's useful is all closed, Right. Although open source is catching up pretty rapidly if you look at.
- [37:47]
Bryce
Yeah, I think everything meta does is open source, right.
- [37:50]
Greg Osuri
Meta, deep seq and all these are like catching up pretty, pretty rapidly. But most usage of AI today is closed source and we don't know what parameters this AI uses. Right. So AI alignment is number one. Understanding these parameters and aligning on these parameters, the system prompts, if you may call it right. I mean, there's a lot of discussion on political biases, a lot of discussion on aspects of the system that is not very transparent. So alignment is like, how can we align these AIs and having a root set of values really? Or what values this AI is built upon. It could be like, hey, don't kill humans, right? Like, okay, what does that mean? In a defensive scenario where you have bad actors and AI actually defending, does that align with don't kill humans? No. So alignment is not as simple as don't do bad things because what's good and what's bad is very, in a way subjective, unfortunately to the system. Is killing a bad thing? Well, depends on one situation, right? So it's like, I think that's the whole alignment discussion on can we create a system where we can be transparent at the same time not disclose our core of the system? AI safety advocates. Now there are obviously two camps of people. One camp of people that think AI is going to kill us all, another camp of people that think that AI is not going to kill us all because it needs humans to actually function. Because in a pure AI economy is kind of useless without a human being. Like if you think of a world where you're just going to put a bunch of AIs in the desert and have their own economy running, well, ultimately it's going to lose its purpose because its purpose is to serve and augment human beings. Right. So I'm of a camp that, that we humans will evolve. Right? We, I think we're, we don't really talk about how we are kind of living in a hyper connected world. I mean we cannot live without our phone, I mean with our cell phones for a minute. Like all aspects of a life. Like if you are cyborgs, basically, cyborgs and I have a sensors in my, my, my arm that'll detect my glucose levels and you know that we are connected 24 bars 7 right. So now with the neural links and a whole lot of devices, now they're going to be permanently placed. And the question you should ask is like, what difference does it make? As we evolve and we become humans or we become this new species, will we act against our own self interest? Right. So I tend to be on the more optimistic trend and be like, no, AI is not going to be killing is and if you put too many safeguards, you're going to just impact the growth of AI. But other people think otherwise. So it's, that's the whole like safety argument on AI.
- [41:12]
Bryce
Yeah, makes sense. Yeah. It's something that I never really dove into, but love the way you explained it and I kind of wanted to also, you know, since this is a crypto podcast, definitely want to bring it back to crypto as well. Just in terms of, you know, I hear this argument of AI and crypto are gonna really get married one day insofar as AIs are going to be using cryptocurrency as a medium of exchange, whether they need to access certain services and they're not going to have a bank account, they're going to have crypto. What do you kind of make of that whole argument?
- [41:46]
Greg Osuri
I mean there's some substance to it, I think. But a stronger align, stronger aspect of, of AI and crypto is incentives. Right? Incentives driving behavior. Sure. AI agents interacting with each other. They're not going to have a bank account but they're also not going to be fully autonomous. I mean let's be real, right? Yeah.
- [42:08]
Bryce
I would never give an AI full access to my wallet.
- [42:11]
Greg Osuri
I mean you can create a separate wallet for it and give it scoped access to it. But an agent, I believe in my life, I use a lot of agents in my home. Agent will act on my behalf. So if it needs a bank account, it's going to use my bank account. Right. Needs a credit card, it's going to have access to my credit card, going to make payments using my credit card. So the thesis around, oh, crypto is the only savior. It's not, let's be real, it's not, it's not a big deal. Sure it'd be nice, can have wallet and pay using bank accounts. But autonomous agents are, you know, I, I'm a firm believer that, you know, autonomous agents will always need a human being to it represents. It's not going to be as autonomous as people imagine would be. And if it is of autonomous, we can imagine there's going to be legislation. Legislation just like how you have LLCs, DAO LLCs today where you can get a bank account if you're a dao. The whole concept of DAO is like you're not going to have a bank account so you need like crypto. But that's change in Wyoming and places. People are going to catch up, legislation going to catch up. UAE is already pretty advanced in terms of I think AI being represented as, as with an idea and whatnot. Right. So a lot of like, I mean I think law and legislation and regulation is going to catch up to this agentic world that we're going towards. And the whole premise of like cryptos, I mean AI is not getting bank account is not very strong. A stronger argument I see is where crypto is really good at is optimizing arcane industries like Deep End. If you're looking at what helium is doing, what Akash be doing. Akash, if you remember in 2023 or 22 there was enormous GPU shortage. CHATGPT came. We basically opened up AI for everyone. And now every company wanted to get GPUs and it was impossible to get GPUs on Amazon. And we saw the trend way early. And we were responsible for creating the trend in crypto because Akash was the only place you can get gpus. And that put us on a map. And now we are going. Fortunately, the GPU crisis has kind of been solved. Not really, but kind of. When I say kind of being solved in the sense there are chips now that are available, you can get Data centers with 5,000 chips, maybe 10,000 if you're lucky. But again, anything beyond that is hard because why energy? Because there are no data centers that can house over 10,000 chips anymore because 10,000 chips miss 10 megawatts. So now you have supply in a lot of small data centers. Now the question is how do you coordinate across these data centers? And that's where crypto comes into play too, right? So we started with solving the energy crisis or the GPU shortage. Now we're solving the coordination crisis because there are too many of these new clouds, you call them Neo clouds, are not quite the hyperscalers, but small companies that can get access to, if you're smart, if you're Wall street type, you know where the data centers are. If you can make a deal to get access to, let's say, you know, 10 megawatt capacity, you know, there are opportunities to get here and there to get 10 megawatt burning coal or whatever, they put a data center there, right? So, so we have a lot of these small guys now. We need coordination across these, these problems. And the third problem now we're entering is like at some point we're going to run out of the 10 megawatts too. And we are running out right now. Now the energy, right? Like, okay, if you have distributed energy grids, essentially distributed energy power plants, 30, 50 kilowatts to 100 kilowatt power plants, how do you coordinate, how do you finance, how do you bootstrap this industry? Now what Daylight and others are doing, it's phenomenal, right? They're, they are bootstrapping essentially solar energy, capturing power plants. And if we can move compute over there and create an incentive layer for you to participate in this new era of decentralized AI, which I believe will take off mid year, like June, July, time frame. Why I'm pretty confident on the timeframe is I'm looking at the progress these companies are making. By the way, I'm heavily involved in a lot of these companies either through investment or a lot of them actually use Akash, like News Research and paralysis and Prime Intellect and all of them actually use Akash right now to do their research. So I keep up quite a lot and we also spend quite a lot of time in the AI conferences. And ICML is in June and there are enormous amounts of papers that are being published by decentralized AI companies. That was not the case about a year ago. So there's a lot of mainstream AI talent that's coming into decentralized AI because we are seeing this progress. So I think there's going to be an explosion of decentralized AI beginning with the training and then we're going to go to inference. And now the question is who's going to power this new revolution? Who's going to provide the chips, the energy? I think that'll open up the deep end aspect of it, including bandwidth, as a matter of fact. Right. So I mean if you have remote clusters, you definitely need bandwidth to connect them to remote clusters. So every layer of AI will be decentralized. And we have an enormous opportunity right now to decentralize using deep end because helium is doing phenomenally well with their bandwidth marketplace. Right. And we are also opening up, fortunately Starlink is opening up, increasing their capacity too. So we have designed a prototype on how a data center should look like. It's on Akash. If you look at Google Akash at home, you'll be able to see a model spec. We wrote out as to what could a data center look like in terms of redundant ISPs, in terms of redundant energy sources, be it solar or be it, you know, how do you, you know, or be generators. And there's another big problem with generators by the way. That's five year lag time to get industrial grid generators.
- [48:43]
Bryce
Wow.
- [48:43]
Greg Osuri
So I mean smaller generators are easier to get than industrial grade generators. So if you, if someone comes to me with a data center plan and be like, where are you going to get your generators from? I don't know how. I have to look into a little bit more deeper as to why we have this like five year lag time. Some blame China.
- [49:05]
Bryce
I was gonna say China must be China.
- [49:06]
Greg Osuri
But it is a supply chain problem for sure. I mean some blame the consensus about COVID really messed up quite a lot. We're still, you know, getting impacted with the supply chain disruptions Covid caused. Yeah, during COVID because generator industry took a huge, huge hit. And a lot of that is because of Chinese lockdowns. If you remember, a lot of the manufacturing was in China. In China were very strict in terms of how they restricted access to supply chain. Right. I mean access to factories and whatnot. So that we're still recovering from that damage and we can go on about like manufacturing independence and things. The critical aspect of being able to produce generators and batteries as well. Right. So anyway, so let's say that problem gets solved another five years.
- [50:00]
Bryce
I know Trump's working on it.
- [50:03]
Greg Osuri
There are a lot of aspects. I mean, it's very delicate situation we have with rare earth minerals right now because. Well, on one aspect, we need the AS rail to produce batteries, right? Yep. Lithium, cobalt and nickel, sorry, nickel, Lithium, cobalt and cobalt, which consumes China controls 95% of them. But there's Greenland.
- [50:24]
Bryce
I was going to say Greenland is probably why this is now on the hotspot, because I saw a map of Greenland. It's like all rare earth minerals.
- [50:30]
Greg Osuri
Yes. That's the primary reason why we have to liberate Greenland. Liberate from their own stupidity. Anyways. Is Europe not mining these things. Yeah. There's a deal with Ukraine that hopefully will go through. Hopefully. Because they have quite a lot of rare earth minerals. Russia has. There are. I mean, India found some rare earth minerals as well. I mean these take time to, to come out. Right. As. As they. As a world is hopefully going towards a post Chinese dominance when it comes to these aspects. Because whatever reason we let. I mean, I talked to have friends in the DOD and the State Department that spend quite a lot of time in Africa because that's where even though China doesn't produce these rare earth minerals, by the way, they come from Africa and they come from Latin America. They just bought up all the companies and they, they strategically over the last.
- [51:23]
Bryce
10 years, the Silk Road Initiative.
- [51:26]
Greg Osuri
The Silk Road. The BRI Belt Road Initiative. The Silk Road. Right. The reconstruction of the Silk Road. The BRI initiative or the Belt Road Initiative really bought up the ports and bought up the transportation infrastructure and bought up the mining infrastructure in primarily in Africa and resource rich countries by giving them loans and by national, by, you know, basically like taking over the ports when they don't pay back the loans. Right, yeah.
- [51:58]
Bryce
Predatory loans for sure.
- [51:59]
Greg Osuri
It's crazy how much. And I have people and I don't have friends that are in Africa. I mean, there are places where even though they could secure the rare minerals, there's no way to transport them to us. So it's a very, very challenging situation. I think what the current administration, what Trump is doing is really to weaken China is not. The tariffs are not. These are not tariffs. These are embargo, essentially. This is not a tariff. Anything about 100% is an embargo. It's Not a tariff. An embargo is an act of war. Right. I mean, we are going very aggressively and I think a big reason is to, to secure our independence when it comes to minerals, even nuclear uranium, because we need energy. Right. There's quite a lot of. There's a big fight there.
- [52:50]
Bryce
Yeah, it's really just like, you know, if we play out or game out all these different scenarios where we might be in a kinetic war, it's like you start to look at the supply chain, you're like, wow, we are dependent on our adversary for pretty much everything crucial. So how do. But, you know, like, how do we. Again, I'm. I'm glad I'm not the one tasked with figuring out how to get us out of this, you know, conundrum. But it is, it is quite an intertwined mess.
- [53:16]
Greg Osuri
I mean, there's signals and you can look at all the. People are not going to say this outright, but if you look at like the trends and what Trump is doing and really to, to get the independence right. I mean, pretty sure Elon Musk is in the center of it because he has a battery factory. He produces batteries. I mean, anyone that understands supply chain, you should be the one, right? So there's a lot of indication on. Apparently there's a new deal with India that was announced yesterday that they're going to mine these rare earth minerals and bring them. India found a huge deposit near Kashmir. Kashmir now tends to be the conflict zone, obviously, right. So it's been in the conflict zone for a while, but now it's going to get worse because the rate of minerals, right. So a lot of, A lot of geopolitical. I mean, follow this rate of minerals, like, you'll find fascinating you a lot of answers you'll find for the questions you may have with the current administration's activities.
- [54:13]
Bryce
Yeah, it all comes back to the basics, right? The rare earth minerals, the things that are actually constructing everything, you know, worth. Worth anything.
- [54:25]
Greg Osuri
Need them for chips, right? Like, you can't produce, like, you can't produce chips. We can't, we can't. If you can't produce chips, you can't have different.
- [54:32]
Bryce
Can't produce weaponry, you can't produce machinery, you can't produce pharmaceuticals. I mean, it's, it's literally the backbone of society now is these, these chips. But I mean, I, I was going to ask like, kind of in closing. You know, it's not a nice question, but it's like kind of like along all this stuff is like what keeps you up at night with crypto and AI. But maybe it's just like, what are you on the lookout for in the future with really paying attention to this one thing here.
- [54:59]
Greg Osuri
I worry about how delicate our grid is. I'm talking about compute grid. Right. Look at large data centers. I mean, hyperscalers, they're only like 24, 30, maybe less than 30 large data centers in America. Right. Like Googles of the world. In case of a conflict. One of the things that's been keeping me up at night really lately has been a conflict scenario potentially in China. Like you look at history, Japan got aggressive during World War II because they were cornered. Like there was a. Tariffs are not new. Tariffs have been in the 80s, in the 1800s, late 1800s, we went on this massive, like, are we going to do this tariffs and whatnot. And there was emphasis on cutting off Japan. And Japan being cut off started their imperialistic agenda, you know, invading Manchurian, the Chinese aspect. And they actually first invaded Korea and then they. To Manchurian. They had to resort because they couldn't get. They couldn't get oil because America cut off oil. Right. And if you.
- [56:09]
Bryce
That was kind of why they bombed Pearl harbor is to like, basically.
- [56:11]
Greg Osuri
Well, that eventually came. Pearl harbor came eventually because Pearl harbor was a response or a signal sent to us to stop meddling in Asia because Japan started invading Korea and then eventually China because they needed resources they didn't have. And us started funding Chinese independent movement, like with the. With the. The Nationalists and the Communists. Right. With Mao and Ching Ba Shek. And obviously Japan didn't like it. Japan wanted them not to fund like we were doing.
- [56:47]
Bryce
I can't believe America funded China like that.
- [56:50]
Greg Osuri
It did. It did. Wow. Funded.
- [56:52]
Bryce
I need to take more history lessons.
- [56:54]
Greg Osuri
And then Japan responded by bombing Pearl harbor as a way to stop America to fund. Right. So one after another. So World War II started for a lot less than what we have right now in terms of trade embargo. That's why, you know, you see trade embargo, it is an act of war. There's no way to look at it. So if you corner China, China being a massive power, is going to explode and invade Taiwan and do things that it. It needs to do because they run out of options. Right. So that's why we need a deal with China. No one wants a war with China because that's. No one's going to win. Win a war, Right?
- [57:30]
Bryce
Yeah.
- [57:31]
Greg Osuri
So in a conflict scenario, I think before Ian, we go into physical bombing stuff, I think it's going to be infrastructure that's targeted. Right. In every war, you go after communications, you go after AI go after. So we have 30 data centers. We all know where these data centers are. Unite. It's a public knowledge. What will it take for an adversary to attack these data centers?
- [57:57]
Bryce
And with the new hacking, it's all going to be hacking. Right?
- [58:00]
Greg Osuri
But hacking, you have, you have drones. You have, you know like a million drones that can be 3D printed these days. It's not. This is. The new warfare is not going to be a traditional big bombers on a, on a, you know, aircraft carrier coming to your shore and bombing. It's going to be a small, relentless, like ruthless, like very, very shadow undetectable stuff, right?
- [58:25]
Bryce
Yes. Swarms.
- [58:26]
Greg Osuri
That's the kind of stuff that swarms, like thousand drone swarms. The bombs. How are you going to stop them? We don't produce. America doesn't have the production capability to produce these drones. I think Elon Musk was tweeting about too, like country that can produce drones will be the winner in the next war. It's not going to be nuclear bombs. No one's going there. I mean, it's arcane anyway. The thing that worries me is like, and our electric grid is so weak, our utilities are so weak. And we have to. The way we solve that is to where we have heavily centered around these small areas to decentralize them in a way that it cannot be attacked. Right. Instead of having one large gigawatt data center, how about thousand megawatt data centers?
- [59:18]
Bryce
It's like the Internet. Right. With TCP IP getting started in this decentralized way just in case communications broke down in World War II. Let's randomize it, send it all these different areas and decentralize it.
- [59:30]
Greg Osuri
Exactly. So that's the kind of stuff that keeps me up at night. The challenge of decentralizing is it's not that easy. Right. And we can talk about all the concepts. Yeah. Bitcoin is great because bitcoin is one operation like writing, putting in and putting out. Like writing and reading from a database. Yeah. Decentralized data center. Well, I mean there are a million problems in a data center when we have data centers on network. I know how the frequency at which communication happens. Not easy. It's very challenging. But fortunately with AI you talk about vibe coding, you can also do Vibe ops. Right. And this is something we're working at Akash too. I mean we have been heavily researching on DevOps is one of those things that I've been doing DevOps for about 15 years now. DevOps is one of those things that is heavily optimized for automation and also heavily repetitive. And when you have automation, optimizable repetitive tasks, that's number one candidate for AI. And we look at how can we achieve about 10,000 of these small modular data centers connected to AI, connected to Akash and essentially having a truly decentralized infrastructure that is attack proof, that is if one node goes down or cluster nodes go down, you're not going to take that infrastructure. And I think I live in Texas. I moved from San Francisco, living in San Francisco for 15 years now. And one of the reasons is I want to have enough space so I can build off the grid home where my family is protected regardless of what happens to the grid. And I mean I am optimistic at the same time pessimistic.
- [61:19]
Bryce
Have you read the book the Spider and the Starfish?
- [61:22]
Greg Osuri
No, I don't believe I did.
- [61:24]
Bryce
It's a good one. It's an easy read. It's all about these different decentralization success stories of the past. But it also talks about how decentralization thrives in nature and different structures and like that.
- [61:40]
Greg Osuri
So I love that.
- [61:41]
Bryce
Yeah, check it out. It was, it was recommended to us by one of our guests many years ago. So I read that one and think you'd enjoy it. But Greg, look, I mean I truly, truly, I don't say this to every guest but we'd love to have you back on sometime in the near future. I thought this was a really enlightening conversation. I love the way that it unfolded. Did it was, it was different. I thought it was really invigorating. So we appreciate you guiding us along this, this AI journey.
- [62:06]
Greg Osuri
Awesome. Thanks.
- [62:08]
Bryce
Absolutely. So everybody who's listening, that was Greg Osuri who's the CEO and founder of Akash Network. Thank you so much for joining us. Come back next week, we'll have some more great guests for you and Greg. Take.
- [62:25]
Greg Osuri
Foreign.
- [62:32]
Shopify Representative
What'S the difference between DIY and doing it yourself? It's the difference between a part time passion and a full time business. WIX gives you the power to turn your passion into a money maker with a website that fits your unique vision. Let your ideas flow with AI that guides you but keeps you in the driver's seat. Manage your business from one dashboard and keep it growing with built in marketing features. It's time to turn your daydream into your dream job. Go to wix.com@ritual. We're celebrating all moms this Mother's Day. Current and aspiring. So whether you're looking for fertility support, Natal Coline or our best selling clinically backed prenatal. We've got you covered. Find your ritual go to ritual.com podcast and for a limited time get 40% off your first month. These statements have not been evaluated by the Food and Drug Administration. This product is not intended to diagnose, treat, cure or prevent any disease. Based on net sales from industry Data including Nielsen IQ, Stackline Enumerator as of Q4 2024 and top 5 products.